Model-Based Approaches to Channel Charting

TL;DR

This paper introduces new model-based algorithms for channel charting that outperform existing methods in estimating user location parameters, with some approaches suitable for efficient hardware implementation.

Contribution

The paper proposes novel model-based algorithms for channel charting, including ISQ, LR, MM, and RS, demonstrating improved performance over traditional methods like PCA, SM, and AE.

Findings

ISQ, LR, and MM outperform PCA, SM, and AE in accuracy.

MM slightly outperforms the JM algorithm with higher complexity.

RS offers similar performance to MM and JM but is more hardware-friendly.

Abstract

We present new ways of producing a channel chart [1] employing model-based approaches. We estimate the angle of arrival theta and the distance rho between the base station and the user equipment by employing our algorithms, inverse of the root sum squares of channel coefficients (ISQ) algorithm, linear regression (LR) algorithm, and MUSIC/MUSIC (MM) algorithm. We compare these methods with the channel charting algorithms principal component analysis (PCA), Sammon's method (SM), and autoencoder (AE) [1]. We show that ISQ, LR, and MM surpass PCA, SM, and AE in performance. We also compare our algorithm MM with an algorithm from the literature that uses the MUSIC algorithm jointly on theta and rho. We call this algorithm the JM algorithm. JM performs very slightly better than MM but at a substantial increase in complexity. Finally, we introduce the rotate-and-sum (RS) algorithm which has about the same performance as the MM and JM algorithms. Unlike MUSIC, RS does not employ eigenvalue and eigenvector analysis. Thus, it is more suitable for direct register transfer logic (RTL) implementation.